This Year’s Best in Hematology Diagnosis: Mapping the Heterogeneity of Chronic Lymphocytic Leukemia

The names we give hematologic diseases tend to be descriptive in nature and help both the physician and the patient conceptualize each specific affliction. However, simple disease names, such as “chronic lymphocytic leukemia” (CLL) belie the intrinsic biologic heterogeneity that exists within each hematologic neoplasm. As the term “chronic” implies, CLL typically displays an indolent behavior, and many patients may never need treatment. However, patient outcomes are highly variable; clinical staging systems such as the Rai and Binet classifications were proposed decades ago to predict this variable patient prognosis.^1,2  For example, the Binet system studied 295 patients with CLL and used blood counts and lymphadenopathy extent to divide patients into three groups, with median survivals ranging from only two years to the same as that of age-matched controls.²  In the 50 years since the advent of these staging systems, numerous other parameters, including immunophenotype, cytogenetic aberrations, mutation signatures, epigenetic signatures, somatic hypermutation of the immunoglobulin heavy locus variable (IGHV) regions, rapid lymphocyte doubling time, and biomarkers of rapid cell turnover have all been identified as prognostic factors in CLL. However, as with many other hematologic neoplasms, it has been uncertain as to how we can integrate data from these various prognostic schemes to inform patient management. Moreover, it is often challenging to interpret mixed messages when one factor (such as highly mutated IGHV status) may suggest a favorable prognosis, while another factor (such as deletion of the TP53 locus at 17p13) suggests an adverse prognosis.

Dr. Binyamin A. Knisbacher and colleagues interrogated samples from 1,095 patients with CLL with a multi-omic approach that included assessment of mutations by next-generation sequencing (whole-exome or whole-genome), expression profiling by RNA-sequencing (RNA-seq), and DNA methylation epigenetic profiling. Although each of these approaches has been used in the past to analyze CLL samples, this study represented the largest such dataset to date to be examined and was unique in terms of its simultaneous assessment of somatic mutations, expression profile, and the DNA methylome. They found 82 putative CLL driver genes, including 37 novel drivers that had not been previously associated with CLL. These novel driver genes included those involving pathways already known to be perturbed in CLL such as ERK signaling, DNA damage, and chromatin modification, but they also uncovered novel mechanisms of altered cellular machinery such as those affecting ribosome function (through deletion of key genes at the 5q locus) and mitochondrial function, and structural proteins involved in the cytoskeleton and extracellular matrix. Although these newly identified drivers were relatively infrequent, owing to the intrinsic mutational complexity, almost one-quarter of patients with CLL harbored at least one driver that was newly identified in the study.

The authors also discovered key differences between the two main somatic hypermutation CLL subgroups — those with mutated (M-CLL) and unmutated (U-CLL) IGHV genes. Overall, the types of driver mutations and copy number abnormalities differed between M-CLL and U-CLL, with more driver mutations in the latter. M-CLL and U-CLL also bore unique patterns of gene rearrangements, likely reflecting their different stages of B-cell development. Aberrant V(D) J recombination characterized M-CLL, while class-switch recombination characterized U-CLL. Finally, the inferred timing of genetic events, such as trisomy 12 and MYD88 mutation, differed in the two molecular CLL subtypes. Taken together, these results indicate that unique genetic mechanisms are inherent to mutated versus unmutated IGHV CLL subtypes, likely explaining their divergent clinical behaviors.

In addition to delineating unique and novel genetic events hard wired into CLL genomes, the authors identified eight distinct gene expression profile clusters, which segregated within U-CLL or M-CLL subtypes and were also associated with specific genetic drivers. Analogous to Philadelphia-like B-cell acute lymphoblastic leukemia (B-ALL), which was initially identified based on a shared expression profile with Philadelphia chromosome–positive B-ALL,³  the authors discovered “phenocopies” of certain genetic CLL subtypes such as cases with an expression signature similar to trisomy 12, despite lacking this cytogenetic abnormality. These data indicate that, despite not currently being used in routine clinical diagnosis, gene expression profiling complements and enhances the information provided by cytogenetics and next-generation sequencing results in CLL diagnostics. These expression classes were shown to be generally stable throughout the course of disease, demonstrating that they are robust and reliable subclassifiers. In a multivariable model, the authors integrated the M-CLL versus U-CLL status with genetic alterations, epigenetic subtypes, and expression clusters to predict patient outcomes (failure-free and overall survival). Not surprisingly, they found that genetics, epigenetics, and gene expression all independently contributed to the outcome of previously untreated patients with CLL in this integrated model (Figure).

The power of the study by Dr. Knisbacher and colleagues is in its examination of a large patient cohort and its multidimensional study of each sample, as correlated with patient outcomes. The large sample size allowed the investigators to identify novel infrequent genetic drivers that were missed in prior, more limited analyses. This paper deserves a Year’s Best recognition because it heralds a new paradigm of disease prognostication. In-depth interrogation of large patient datasets on multiple levels (DNA changes, RNA expression patterns and epigenetic signatures) will increasingly inform prognostic modelling to more accurately guide patient management in the future. This has already been shown to be the case for other diseases such as myelodysplastic syndrome, in which gene expression profiling improved outcome prediction beyond genetic and clinical data.⁴  The main challenge to this approach is the lack of availability of gene expression and epigenetic profiling in clinical practice. The study of Dr. Knisbacher and colleagues is a call to action to expand and diversify our diagnostic armamentarium by including new technologies that have the potential to guide treatment decisions, better predict patient responses to existing therapies, and inform the development of new therapeutic approaches.